Abstract

Stress analysis is one of the most important processes in designing an underground excavation. With the usage of numerical methods on a computer, such as the Boundary Element Method (BEM), the process of stress analysis can be made accurate. However, the computational implementation of stress analysis often requires considerable time and computational resources. For example, in the implementation of BEM, the finer the computational grid is, the longer time will take to compute the results. Based on the research of GPU-accelerated stress analysis in geomechanics (Zsaki, 2011), this thesis investigates one type of acceleration method, which used the parallel computing ability of modern graphics processing units (GPUs), for application to the traditional BEM algorithm. In this thesis, OpenCL was used as the framework to compile and execute programs on GPUs. By transferring and executing the most computational expensive parts of the traditional BEM code onto GPU, a respectable acceleration was achieved. Subsequently, with the application of a two-dimensional circular excavation example, the accuracy of the BEM algorithm implementation on a GPU was verified for both single-precision and double-precision calculations. In addition, two more excavation examples were taken into consideration to assess the accuracy and reduction in solution time. The performance for these three examples successfully verified the GPU-accelerating method and displayed an impressive acceleration effect with the speedup ratio of about 500 for single-precision and 15 for double-precision.